Large Language Models: From Theory to Practice

A hands-on course that teaches how modern AI and large language models work, how to use them effectively for research, and how to navigate their capabilities and limitations responsibly. Below is the detailed course agenda:

Day 1 – Foundations: AI, GenAI, and Modern LLMs 

We ground everyone in what AI, Generative AI, and AGI actually mean, then build a practical mental model for how today’s LLMs work—and where they break. Through clear, visual explanations of transformers, attention, embeddings, tokenization, and MoE architectures, we connect mechanics to behavior (reasoning, “memory,” and the so-called black box). We’ll demystify interpretability efforts, set realistic capability limits, and show how prompting and context engineering change outcomes. By the end of the day, participants can explain how LLMs produce text, diagnose common failure modes, and apply prompting patterns (zero-/few-shot, chain-of-thought) with intention—not superstition. 

Day 2 – Practice: Tools, Research Workflows, and Projects 

We compare leading models and platforms (e.g., GPT, Claude, Gemini) by strengths, guardrails, and fit-for-purpose, then move from demos to doing: using LLMs for quantitative and qualitative research, handling data responsibly, and documenting runs for replicability and auditability. We close with a collaborative project clinic where participants pressure-test ideas, map LLM methods to their research questions, and leave with next-step plans. 

Prerequisites

Participants should bring their notebook and it would be good to have either an OpenAI, Anthropic or Google account (to use ChatGPT, Anthropic and/or Gemini during the course). 

Date & Time

The course will be held at GIGA on 28-29 October 2025, from 10:00-17:00 (on both days). The main target group of this course are GIGA researchers. External participants can also register.

About the trainer

Eduardo Tamaki is a Doctoral Researcher at the GIGA and the Willy Brandt School of Public Policy, University of Erfurt. His research interests include political behavior, public opinion research, and the expanding applications of AI in social science.  He has taught multiple methods courses with a focus on data analysis with R and practical applications of AI in political research.